murmur
A compiler for a C-based SPMD language that generates high-performance SIMD code for CPUs and GPUs.
What is murmur?
Intel® ISPC is a compiler for a C-based SPMD programming language that generates optimized SIMD code for CPU and GPU vector units. It solves the problem of writing high-performance parallel code by abstracting away the complexity of low-level intrinsics, allowing developers to write programs that scale across multiple cores and wide vector architectures. The compiler frequently delivers 3x to 6x speedups on modern hardware compared to scalar code.
Performance-oriented programmers and developers working on computationally intensive applications such as graphics, simulation, scientific computing, or game engines who need to harness SIMD parallelism on CPUs or GPUs.
Developers choose ISPC because it provides near hand-tuned intrinsics performance with much higher programmer productivity, thanks to its familiar C-like syntax and seamless interoperability with existing C/C++ codebases. Its cross-platform support and ability to target a wide range of CPU and GPU architectures make it a versatile tool for performance portability.
Overview
Intel® Implicit SPMD Program Compiler
Use Cases
Best For
- Accelerating graphics rendering pipelines with SIMD optimizations
- Writing high-performance numerical simulations for scientific computing
- Optimizing game engine components like physics or AI computations
- Developing cross-platform compute kernels that target both CPU and GPU
- Porting legacy C code to take advantage of modern vector instructions
- Creating performance-portable libraries that run efficiently on x86 and ARM architectures
Not Ideal For
- Projects with highly irregular or data-dependent parallelism where the SPMD model struggles
- Teams needing extensive pre-built parallel libraries without modifying C/C++ build systems
- Environments with strict compiler toolchain restrictions, such as embedded systems using proprietary compilers
Pros & Cons
Pros
Uses familiar C syntax and idioms, easing adoption for C/C++ developers while adding SPMD extensions, as highlighted in the README's emphasis on programmer productivity.
Delivers code with performance close to hand-written intrinsics, frequently providing 3x to 6x speedups on vector architectures like AVX and NEON, according to the README.
Functions and data structures interoperate directly with C/C++ code without copying, enabling lightweight integration into existing software systems, as stated in the features.
Supports x86, ARM NEON, and Intel GPU architectures across Windows, macOS, and Linux, allowing performance portability over a decade of CPU generations.
Cons
Requires manual integration into C/C++ build pipelines and depends on LLVM, complicating setup compared to standard compilers, especially for teams unfamiliar with custom toolchains.
As a niche compiler, it lacks the extensive libraries, debugging tools, and community resources of mainstream languages, which can hinder development beyond core compute kernels.
The SPMD programming model may not efficiently handle dynamic parallelism or complex control flow, limiting its applicability to data-parallel workloads with uniform execution paths.
Frequently Asked Questions
Related Projects
libcrcMurmur3 hash in C
t1haOne of the fastest hash functions.
jwHashSimple hash table implementation for C.
xxHashExtremely fast non-cryptographic hash algorithm. 2-clause BSD
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.